TNscope: Accurate Detection of Somatic Mutations with Haplotype-based Variant Candidate Detection and Machine Learning Filtering
Abstract
Detection of somatic mutations in tumor samples is important in the clinic, where treatment decisions are increasingly based upon molecular diagnostics. However, accurate detection of these mutations is difficult, due in part to intra-tumor heterogeneity, contamination of the tumor sample with normal tissue and pervasive structural variation. Here, we describe Sentieon TNscope, a haplotype-based somatic variant caller with increased accuracy relative to existing methods. An early engineering version of TNscope was used in our submission to the most recent ICGC-DREAM Somatic Mutation calling challenge. In that challenge, TNscope is the leader in accuracy for SNVs, indels and SVs. To further improve variant calling accuracy, we combined the improvements in the variant caller with machine learning. We benchmarked TNscope using in-silico mixtures of well-characterized Genome in a Bottle (GIAB) samples. TNscope displays higher accuracy than the other benchmarked tools and the accuracy is substantially improved by the machine learning model.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 63
- Assay
- Research Focus
- Approach
Abstract summary
Claude Opus 5.5 · xhigh effort · 18 Oct 2026 · from the abstract · CC BY-NC-ND 4.0
A haplotype-based somatic variant caller paired with a machine learning filter, reported as more accurate than other callers on in-silico mixtures of GIAB reference samples.
Somatic calling in tumors has to cope with subclonal variants, normal-tissue contamination of the sample and widespread structural variation, all of which blur the signal that clinical molecular diagnostics depend on. TNscope assembles candidate variants from haplotypes and then filters them with a machine learning model, which the authors present as more accurate than the callers they compare against.
Key results
ClaimedOn a benchmark built by mixing well-characterized Genome in a Bottle (GIAB) samples in silico, TNscope is reported to be more accurate than the other callers tested.
Stated as higher accuracy than the other benchmarked tools; the abstract names no comparator tool, no mixture fractions and no sensitivity, precision or F-score values. · Abstract
ClaimedThe machine learning filter is credited with a large part of the accuracy, on top of the haplotype-based caller itself.
Described as a substantial improvement from the machine learning model; the model type, its features, its training data and the size of the gain are not given in the abstract. · Abstract
ClaimedAn earlier engineering build of the caller was entered in the ICGC-DREAM somatic mutation calling challenge, where the authors place it first for accuracy on single-nucleotide variants, indels and structural variants.
Reported for the most recent such challenge at the time of writing; no challenge scores, ranking table or round identifier is given, and the submitted version differs from the one described here. · Abstract
Methods, data and reproducibility
Sentieon TNscope calls somatic variants by assembling candidate haplotypes from tumor data and then applies a machine learning model to filter the candidates; accuracy was assessed on tumor-like mixtures created computationally from GIAB reference samples, and an earlier build was run on an ICGC-DREAM challenge dataset.
- in-silico mixtures of well-characterized Genome in a Bottle (GIAB) reference samples, used as a somatic calling benchmark · human
- data from the ICGC-DREAM Somatic Mutation Calling Challenge, used by an early engineering version of the caller
- Methods & tools
- Sentieon TNscope, Genome in a Bottle (GIAB)
Limitations (2)
- No accuracy figures and no named comparator callers appear in the available text, and the benchmark was assembled and run by the authors of the tool being evaluated. · noted in this summary
- Accuracy is measured on computationally mixed reference genomes rather than real tumors, so subclonal structure, purity and structural variation in clinical samples are only approximated. · noted in this summary
Who should read:Read this if you call somatic variants in tumor-normal data and are comparing haplotype-based callers with machine learning filtering against your current pipeline.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- Evaluation of somatic variant calling methods on high coverage tumour-only amplicon sequencing data in a clinical environment
Dhammapal Bharne, Daniel Gaston·bioRxiv·Apr 11, 2026
- Detecting Somatic Mutations Without Matched Normal Samples Using Long Reads
Jared T. Simpson·bioRxiv·Feb 26, 2024
- Using Machine Learning to Identify True Somatic Variants from Next-Generation Sequencing
Chao Wu, Xiaonan Zhao, … Mahdi Sarmady·Clinical ChemistryPeer-reviewed·Oct 31, 2019
- Lancet: genome-wide somatic variant calling using localized colored DeBruijn graphs
Giuseppe Narzisi, André Corvelo, … Michael C. Zody·Communications BiologyPeer-reviewed·Mar 14, 2018
- Attentive deep learning-based tumor-only somatic mutation classifier achieves high accuracy agnostic of tissue type and capture kit.
Robert T. McLaughlin, Maansi Asthana, … David L. Masica·bioRxiv·Dec 7, 2021
- DeepSomatic: Accurate somatic small variant discovery for multiple sequencing technologies
Jimin Park, Daniel E. Cook, … Kishwar Shafin·Nature BiotechnologyPeer-reviewed·Oct 16, 2025